Tool-Use Agent Evaluation on BFCL Multiturn (OOD) v3 (test)
48Base RateTopoCurate-RL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TopoCurate-RLModel Size=32B, Strategy=RL, Topology Modeling=Enabled2026.03 | 48 | 29.5 | 22.5 | 32.5 | |
| TopoCurate-RL (w/o Topology)Model Size=32B, Strategy=RL, Topology Modeling=Disabled2026.03 | 45 | 24.5 | 16.5 | 26 | |
| TopoCurate-SFTModel Size=32B, Strategy=SFT, Topology Modeling=Enabled2026.03 | 43 | 26.5 | 13.5 | 29.5 | |
| MUA-RLModel Size=32B, Strategy=RL2026.03 | 42 | 21.5 | 20 | 30 | |
| TopoCurate-SFT (w/o Topology)Model Size=32B, Strategy=SFT, Topology Modeling=Disabled2026.03 | 41 | 23.5 | 12.5 | 24.5 | |
| APIGen-MT-SFTModel Size=32B, Strategy=SFT2026.03 | 37.5 | 21.5 | 17.5 | 20.5 | |
| MUA-SFTModel Size=32B, Strategy=SFT2026.03 | 35 | 19.5 | 21 | 28.5 | |
| Qwen3-32B-ThinkingModel Size=32B, Strategy=Thinking2026.03 | 34 | 21 | 15.5 | 24.5 | |
| TopoCurate-RLModel Size=8B, Strategy=RL, Topology Modeling=Enabled2026.03 | 29.5 | 16 | 13.5 | 18 | |
| TopoCurate-SFT (w/o Topology)Model Size=8B, Strategy=SFT, Topology Modeling=Disabled2026.03 | 26.5 | 13.5 | 7.5 | 15 | |
| TopoCurate-SFTModel Size=8B, Strategy=SFT, Topology Modeling=Enabled2026.03 | 26.5 | 12.5 | 9 | 16 | |
| TopoCurate-RL (w/o Topology)Model Size=8B, Strategy=RL, Topology Modeling=Disabled2026.03 | 25.5 | 14.5 | 10.5 | 16 | |
| APIGen-MT-SFTModel Size=8B, Strategy=SFT2026.03 | 24.5 | 14 | 5 | 17 | |
| MUA-SFTModel Size=8B, Strategy=SFT2026.03 | 24 | 10 | 11 | 16.5 | |
| Qwen3-8B-ThinkingModel Size=8B, Strategy=Thinking2026.03 | 23 | 13.5 | 4.5 | 13.5 | |
| MUA-RLModel Size=8B, Strategy=RL2026.03 | 21 | 11 | 11.5 | 15 | |
| Simia-Tau-SFTModel Size=8B, Strategy=SFT2026.03 | 4.5 | 2 | 0.5 | 8 | |
| Simia-Tau-RLModel Size=8B, Strategy=RL2026.03 | 4 | 2.5 | 1.5 | 7.5 |